Machine-learned models for electric vehicle component health monitoring

Abstract

Methods, computing systems, and technology for machine-learned vehicle component health monitoring are presented. An example method may include obtaining operating data describing one or more operational characteristics of a component of a subsystem onboard a vehicle. The example method may include generating, using a component longevity model and based on the operating data, a component longevity value for the component. The example method may include generating, using a vehicle usage model and based on the component longevity value, a prognosis for the component. In the example method, the vehicle usage model may be configured to evaluate the component longevity value based on a usage pattern associated with the vehicle. The example method may include initiating, based on the prognosis, a corrective action to mitigate degradation of the component.

What is claimed is: 
     
       1. A computing system for monitoring an operational health of a vehicle subsystem, the computing system comprising:
 a control circuit configured to:
 obtain operating data describing one or more operational characteristics of a component of a subsystem onboard a vehicle; 
 generate, using a component longevity model and based on the operating data, a component longevity value for the component; 
 generate, using a vehicle usage model and based on the component longevity, value, a prognosis for the component, wherein the vehicle usage model is configured to evaluate the component longevity value based on a usage pattern associated with the vehicle, wherein the vehicle usage model comprises a machine-learned neural network trained to recognize anomalous usage patterns of the vehicle based on the one or more operational characteristics; 
 recognize, using the vehicle usage model, an anomalous usage pattern outside a domain of the component longevity model; 
 override, based on recognizing the anomalous usage pattern, the component longevity value with the prognosis; and 
 initiate, based on the prognosis, a corrective action to mitigate degradation of the component, wherein the corrective action comprises causing the vehicle to change a drive mode. 
 
 
     
     
       2. The computing system of  claim 1 , wherein the prognosis comprises an effective longevity value for the component. 
     
     
       3. The computing system of  claim 2 , wherein the control circuit is configured to:
 obtain an initial effective longevity value; and 
 generate, using the component longevity model and based on the operating data and the initial effective longevity value, the component longevity value for the component. 
 
     
     
       4. The computing system of  claim 3 , wherein the control circuit is configured to:
 iteratively update a storage location that stores the effective longevity value. 
 
     
     
       5. The computing system of  claim 1 , wherein the component longevity model comprises a physics-based model. 
     
     
       6. The computing system of  claim 1 , wherein the component longevity model comprises a machine-learned neural network trained to output an association between a set of input operational characteristics and one or more failure modes of the component. 
     
     
       7. The computing system of  claim 6 , wherein the component is an active electrical component, and wherein the one or more failure modes comprises at least one of the following: (i) bond wire liftoff, (ii) dielectric breakdown, (iii) threshold voltage instability, or (iv) bias temperature instability. 
     
     
       8. The computing system of  claim 1 , wherein the control circuit is configured to:
 for each respective component of a plurality of components of the subsystem:
 obtain respective operating data describing one or more respective operational characteristics of the respective component of the plurality of components; and 
 generate, using a respective component longevity model and based on the respective operating data, a respective component longevity value for the respective component; and 
 
 generate, using the vehicle usage model and based on the plurality of respective component longevity values, a prognosis for the subsystem. 
 
     
     
       9. The computing system of  claim 2 , wherein the vehicle usage model comprises a machine-learned neural network trained to generate the effective longevity value based on a latent embedding of the usage pattern. 
     
     
       10. The computing system of  claim 1 , wherein the control circuit is configured to:
 obtain the usage pattern from a database of usage patterns, the usage pattern stored in the database in association with a particular user of the vehicle. 
 
     
     
       11. The computing system of  claim 1 , wherein initiating, based on the prognosis, the corrective action comprises:
 initiating a control signal configured to cause the vehicle to render a warning message to an occupant of the vehicle. 
 
     
     
       12. The computing system of  claim 1 , wherein initiating, based on the prognosis, the corrective action comprises:
 transmitting, to a remote server, a message indicating the prognosis. 
 
     
     
       13. The computing system of  claim 1 , wherein the corrective action is configured to reduce a load on the component. 
     
     
       14. A method for monitoring an operational health of a vehicle subsystem, the method comprising:
 obtaining operating data describing one or more operational characteristics of a component of a subsystem onboard a vehicle; 
 generating, using a component longevity model and based on the operating data, a component longevity value for the component; 
 generating, using a vehicle usage model and based on the component longevity value, a prognosis for the component, wherein the vehicle usage model is configured to evaluate the component longevity value based on a usage pattern associated with the vehicle, wherein the vehicle usage model comprises a machine-learned neural network trained to recognize anomalous usage patterns of the vehicle based on the one or more operational characteristics; 
 recognizing, using the vehicle usage model, an anomalous usage pattern outside a domain of the component longevity model; 
 overriding, based on recognizing the anomalous usage pattern, the component longevity value with the prognosis; and 
 initiating, based on the prognosis, a corrective action to mitigate degradation of the component, wherein the corrective action comprises causing the vehicle to change a drive mode. 
 
     
     
       15. The method of  claim 14 , wherein the prognosis comprises an effective longevity value for the component, and the method comprises:
 obtaining an initial effective longevity value; and 
 generating, using the component longevity model and based on the operating data and the initial effective longevity value, the component longevity value for the component. 
 
     
     
       16. The method of  claim 14 , wherein the component longevity model comprises a machine learned neural network trained to output an association between a set of input operational characteristics and one or more failure modes of the component. 
     
     
       17. The method of  claim 14 , comprising:
 for each respective component of a plurality of components of the subsystem:
 obtaining respective operating data describing one or more respective operational characteristics of the respective component of the plurality of components; and 
 generating, using a respective component longevity model and based on the respective operating data, a respective component longevity value for the respective component; and 
 
 generating, using the vehicle usage model and based on the plurality of respective component longevity values, a prognosis for the subsystem. 
 
     
     
       18. The method of  claim 14 , comprising:
 obtaining the usage pattern from a database of usage patterns, the usage pattern stored in the database in association with a particular user of the vehicle. 
 
     
     
       19. The method of  claim 14 , wherein the corrective action is configured to reduce a load on the component. 
     
     
       20. One or more non-transitory computer-readable media that store instructions that are executable by a control circuit to:
 obtain operating data describing one or more operational characteristics of a component of a subsystem onboard a vehicle; 
 generate, using a component longevity model and based on the operating data, a component longevity value for the component; 
 generate, using a vehicle usage model and based on the component longevity value, a prognosis for the component, wherein the vehicle usage model is configured to evaluate the component longevity value based on a usage pattern associated with the vehicle; wherein the vehicle usage model comprises a machine-learned neural network trained to recognize anomalous usage patterns of the vehicle based on the one or more operational characteristics; 
 recognize, using the vehicle usage model, an anomalous usage pattern outside a domain of the component longevity model; 
 override, based on recognizing the anomalous usage pattern, the component longevity value with the prognosis; and 
 initiate, based on the prognosis, a corrective action to mitigate degradation of the component, wherein the corrective action comprises causing the vehicle to change a drive mode.